Protocols for agentic AI payments in retail

Agentic AI can now autonomously shop and pay on behalf of consumers. But making this work requires specialized protocols to enable secure payments without human involvement. This is where protocols for AI-driven payments come in: these standards define how data is exchanged between AI agents, merchants, and payment providers for authentication, authorization, and traceability.

Protocols for AI-driven payments are standardized rules that allow AI programs to complete transactions independently. Traditional online payments rely on confirmations such as SMS codes (3D Secure) or passwords – approaches that don’t work for autonomous AI. Instead, these systems store user intent (e.g., “up to €50 for product X”) in digital contracts that can be verified by all parties involved.

Without such protocols, merchant systems often flag AI requests as suspicious bots and automatically block them. Standard security measures like SMS confirmations or password prompts fail because there’s no human present to complete them. At the same time, fraud risks increase: how can you prove that an AI is truly acting on behalf of the correct customer and hasn’t been compromised?

These protocols address several challenges at once. They verify each AI agent’s identity through a digital ID with a transaction history, enforce spending limits, and create auditable records of every transaction. This allows merchants to confidently accept automated purchases – for example, when an AI executes a purchase after a price drop – without fear of unauthorized charges or disputes. For retailers, this could mean higher conversion rates through seamless AI integration, without the usual security friction.

The protocols operate in three main steps:

1. Grant permission
2. Validate the cart
3. Execute the payment

First, the user creates a statement of intent (e.g., “buy product X for up to €50”), which the AI digitally signs and stores. This is then matched against the shopping cart to ensure that price, availability, and conditions align. Only then is the payment sent to the payment system via standardized interfaces (REST APIs).

Merchants must also provide machine-readable catalogs (price, inventory, conditions) via APIs. This is essential for visibility on AI platforms.

Every step remains traceable: the AI’s identity is verified, limits are enforced, and the entire chain – from intent to charge – is logged. Examples such as Google’s AP2 use so-called “mandates” (digital authorizations) that translate user preferences into machine-readable formats. Merchant systems can reliably recognize legitimate AI-driven purchases and process them without manual review.

Although many current use cases are emerging in online retail, agentic AI payments extend far beyond traditional e-commerce. AI agents do not operate only in webshops; they are increasingly active across channels – for example, when they research products, check in-store availability, trigger reservations, or prepare purchases in the background that are later picked up or used in physical retail.

This marks a shift in a core principle of payment: agentic systems integrate payments into systems, not interfaces. In other words, the transaction no longer takes place visibly during the checkout process, but is handled in the background between APIs, platforms, and retail systems – triggered by defined rules and user intentions rather than by active input.

For retailers, this means the decisive factor is no longer the sales channel, but the ability to make transactions accessible to machines at the system level. Agentic payment protocols are therefore becoming an infrastructure issue for all of retail – from the API in the online shop to integration into omnichannel processes and store systems.

For retailers, these protocols open the door to agentic commerce: AI agents can be seamlessly integrated into online stores and complete purchases without users ever leaving the platform – for example, via ChatGPT, Google Gemini, or Perplexity. To participate, merchants need to provide machine-readable product catalogs (pricing, availability, conditions) and checkout processes that AI agents can access and use.

• Higher conversion rates: automated purchases triggered by price drops, back-in-stock alerts, or personalized recommendations
• Fewer drop-offs: no manual confirmation steps, leading to completed carts
• Omnichannel potential: AI connects online and physical retail (e.g., automated in-store reservations)

• Technical and operational complexity: integration, data quality, and lack of standardization
• Risks and regulation: unclear liability, new fraud vectors, legal uncertainty
• Dependency and trust: growing platform power, limited user adoption

Since 2025, major payment providers have been developing protocols for AI-driven transactions – a fast-moving market with multiple players and early merchant integrations.

1. Visa Trusted Agent Protocol (October 2025): Developed with Cloudflare, it enables AI agents to send cryptographically signed HTTP requests so merchants can distinguish legitimate purchases from bots – available via the Visa Developer Center
2. Google’s AP2 protocol (September 2025): More than 60 partners (including PayPal and Mastercard); Wayfair is already using it for AI shopping in Google Search and the Gemini app – enabling direct checkout without leaving the page
3. Mastercard Agent Pay: Focused on secure agent interactions in commerce; In March 2026, Santander and Mastercard announced the first end-to-end payment transaction conducted in real time by an AI agent in Europe

Additional protocols include ACP (OpenAI/Stripe), x402 (Coinbase), and MPP (Stripe/Tempo) – all built on the HTTP 402 standard for interoperability.

The market is fragmented but growing rapidly. Retailers like Wayfair demonstrate that real-world applications are already live, though most remain early-stage use cases without broad scalability. Many projects are still in beta.

A 2025 Mastercard study shows that 39% of Germans are open to using AI agents for shopping – rising to 56% among those aged 30 to 39 – while 26% expect normalization within five years.

Automated purchases could boost conversion rates by removing confirmation steps and making retailers accessible via AI platforms like ChatGPT.

At the same time, significant risks persist. Legal gray areas in payment services regulation (PSD2) and data protection (GDPR) are not well suited to autonomous agents, and liability for erroneous decisions remains unresolved. Fraud risks – such as manipulated AI identities or deepfakes – are increasing, while regulators like BaFin (Germany’s Federal Financial Supervisory Authority) and the European Banking Authority (EBA) are calling for stricter transparency and “human-in-the-loop” requirements. This is not merely a gray area, but an unresolved core issue.

Only a small number of AI agent projects have reached production-level maturity so far, largely due to high compliance hurdles. In addition, European retailers may face disadvantages compared to US tech giants due to growing platform concentration.

In short, protocols for agentic AI payments enable autonomous purchasing but introduce significant challenges in security and regulation. Retailers must decide whether to integrate with AI platforms – gaining the potential for higher conversions while taking on legal and fraud-related risks. For now, the trajectory of this development remains uncertain.